Assumptions and Diagnostics for t-Tests
You'll learn: What conditions must be met for t-test results to be trustworthy, and how to check them.
Why Assumptions Matter
A t-test is like a recipe—it only works well when you have the right ingredients. Even if your calculation is mathematically correct, violating the underlying assumptions can make your conclusions unreliable or completely invalid. Before accepting any t-test result, you must verify that your data meets the necessary conditions.
Core Assumptions for t-Tests
1. Normality
The data (or differences, in paired tests) should come from a normally distributed population. For small samples (n < 30), this matters a lot. For larger samples, the Central Limit Theorem provides some protection against non-normality.
Check it: Use histograms, normal probability plots (Q-Q plots), or formal tests like Shapiro-Wilk. Look for severe skewness or outliers.
2. Independence
Each observation must be independent of others—no observation should influence another. Think of flipping coins: each flip doesn't affect the next.
Check it: Consider your data collection method. Were subjects randomly selected? Is there clustering (like siblings, repeated measures on same units)?
3. Equal Variances (for two-sample tests)
When comparing two groups with a standard two-sample t-test, both populations should have similar variances (homoscedasticity). Welch's test relaxes this assumption.
Check it: Compare sample standard deviations (rule of thumb: ratio shouldn't exceed 2:1), or use Levene's test.
What If Assumptions Fail?
Consider transformations (log, square root), non-parametric alternatives (Wilcoxon tests), or larger samples to invoke the CLT.
Key Takeaway: Always verify normality, independence, and (for two-sample tests) equal variances before trusting t-test conclusions—violations can invalidate your results entirely.